PUB: Product Recommendation with Users' Buying Intents on Microblogs

摘要

Recommendation systems mostly rely on users' purchase records. However, they may suffer problems like "cold-start" because of the lack of users' profiles and products' demographic information. In this paper, we develop a method called PUB, which detects users' buying intents from their own tweets, considers their needs, and extracts their demographic information from their public profiles. We then recommend products for users by constructing a heterogeneous information network including users, products, and attributes of both. In particular, we consider users' shopping psychology, and recommend products that better meet their needs. We conduct extensive experiments on both direct intent recommendation and additional product recommendation. We also figure out users' potential preference which can help to recommend a great varied types of products.